Beam_labels of Waymo and nuScenes datasets

[CBY-9527]

When creating the dataset information, I found that the KITTI dataset uses "python -m pcdet.datasets.kitti.kitti_beam_id --data_path data/kitti". However, there are no similar operations and corresponding py files for the Waymo and nuScenes datasets.

When pre-training Waymo, an error is reported: FileNotFoundError: [Errno 2] No such file or directory: '/media/vision/ycb/data/waymo/ waymo_processed_data_v0_5_0/segment-13145971249179441231_1640_000_1660_000_with_camera_labels/0163.beam_labels.npy'.

If it is convenient for you, please upload "kitti_beam_id.py", "kitti_dataset_mean_teacher.py" and "kitti_dataset_mt.py" (in pcdet/datasets/kitti/) corresponding to waymo and nuScenes. Looking forward for your reply, thank you very much!

[WoodwindHu]

The code for generate the waymo beam id is in waymo_utils.py, while the beam id of nuscenes is calculated in real-time with data_augmentor.py.

[CBY-9527]

The code for generate the waymo beam id is in waymo_utils.py, while the beam id of nuscenes is calculated in real-time with data_augmentor.py.

Thank you very much for your reply. I tried to generate the beam id during Waymo data preprocessing, but it failed because my raw_data was corrupted. Later, I tried to calculate the beam id of Waymo in real-time through data_augmentor.py, but when I used the model trained in the source domain to directly test the nus, the performance was almost zero. Can you give some suggestions, and the random_beam_downsample in data_augmentor.py can be used to correctly calculate the beam id for Waymo in real-time? Thank you so much!

[WoodwindHu]

but when I used the model trained in the source domain to directly test the nus, the performance was almost zero.

Do you mean train in waymo and directly test the nus?

[WoodwindHu]

Can you give some suggestions, and the random_beam_downsample in data_augmentor.py can be used to correctly calculate the beam id for Waymo in real-time?

The beam id of waymo can be directly extracted from the dataset, there is no need to calculate the beam id. You can check waymo_utils.py for detail.